Study On Methods Based On Biological Network For Predicting MiRNA Associated With Complex Disease

MiRNA is an evolutionarily conserved,single-stranded,endogenous non-coding RNA of approximately 20 nucleotides in length in eukaryotes.Although it does not encode proteins,it is involved in the regulation of target genes in cells.It plays a key role in biological processes such as proliferation,growth,and apoptosis,which can help us understand the development of the disease from the molecular perspective.The detection miRNA and its function will help to understand the regulation and control mechanism of miRNA,which is helpful for understanding the mechanism of disease development and is very important for the prevention and treatment of human diseases.In recent years,a large number of miRNA biological data have been continuously formed,and the functions of most miRNAs are still unknown.The correlation between the recognition of miRNAs and diseases is a hot topic in biological research.Researchers can accurately identify the association between miRNAs and diseases through sophisticated biological experiments.However,they require high experimental conditions and have a long cycle.Computational methods infer the association based on known biological data is a best supplement to biological experiment.However,most current methods have the disadvantages of low prediction accuracy,negative sample acquisition,inability to predict isolated diseases and new miRNAs,and poor generalization ability.This paper proposes three miRNA-related disease prediction methods to address these problems.The main research work is as follows:1)Heterogeneous bipartite network link prediction methods based on common neighbors were used to infer miRNA-disease associations.Inspired by the common neighbors of the single-division network,the concept of the common neighborhood of the bipartite network is defined and the definitions of eight common neighbors are given.Through these indicators,the initial association scores of diseases and m iRNA nodes can be calculated,and the initial score can be used to predict Disease-miRNA associations.In order to obtain more accurate prediction results,the similarity between disease and disease and the similarity between miRNAs are introduced.Based o n the initial scores,the secondary prediction scores based on the spatial similarity of disease and the common neighbors based on miRNA similarity are obtained.The common neighbor link prediction algorithm based on miRNA functional similarity greatly improves the prediction accuracy and can predict new miRNAs,but it cannot be used for the prediction of isolated diseases.The common neighbor link prediction algorithm based on the functional similarity of the disease can be applied to isolated disease prediction but no prediction ability for new miRNAs.Finally,we weighted these two algorithms to form an integration algorithm.This method concentrates the advantages of the above algorithms,has high prediction accuracy,and can be used for the prediction of isolated diseases and new miRNAs.The leave-one-out cross-validation performed on the two different datasets of the gold baseline dataset and the predictive dataset shows that the algorithm has good predictive power.In the case study,we conducted a predictive study of breast and colon cancers.Both associated miRNA and isolated disease predictions showed good predictive power,and most of the prediction results were confirmed by the database.2)Global Similarity Method Based on a Two-tier Random Walk for the Prediction of miRNA–Disease Association.Considering that using global information can improve prediction accuracy,we try to use the Laplacian score of the graph to calculate the global similarity of the network by using the optimized disease seeds and miRNA seeds in the miRNA network and the disease network,respectively.Random walk,the Pearson coefficient of each vector of the global similarity matrix between the stable vector and the miRNA,the stable vector obtained from the miRNA network,and the Pearson coefficient of each vector of the global similarity matrix between diseases are taken as the prediction score of miRNAs-disease.Finally,the predicted scores obtained from the two networks are weighted as the final miRNA-disease association prediction score.This method can predict without negative samples.Our method is superior to the existing methods in the prediction accuracy,and there is a clear advantage in the prediction of new miRNAs,particularly solitary diseases.Most of the miRNA-disease associations predicted in the top 50 of the case studies can be experimentally verified,fully demonstrating that this method is feasible and effective.3)Information-diffusion disease-associated miRNA prediction based on network consistency.Reasonable construction of disease and miRNA similarity relationship can improve the prediction accuracy of the calculation method.We first integrate experimentally validated disease and miRNA association information,disease semantic scores,and maps of Laplacian scores to build a global network of disease similarities.Use miRNA family information,miRNA functional similarity,Laplacian score of maps to construct miRNA global similarity network,and then use the known disease and miRNA association information and global similarity between miRNA nodes to build miRNA-based global similarity information disease-miRNA association network ASm;The disease-miRNA association network ASd based on the disease global similarity information was constructed by using the known similarities between disease and miRNA related information disease nodes,and then the network homogeneity diffusion seed was obtained by combining the global similarity network and ASm and ASd,respectively.In the disease global similarity network an d the miRNA global similarity network,the stable diffusion spectrum was randomly walked to obtain the prediction score.Finally,these two prediction scores were weighted to obtain the final miRNA-disease-associated miRNA prediction score.The method does not require negative samples and can be used for the prediction of new miRNAs and isolated diseases,and the algorithm is simple in design.The LOOCV evaluation results on the gold baseline dataset and the prediction dataset show that our method is superio r to the two methods we proposed earlier and methods of others.Most of the top 50 predictions in the case studies were confirmed by the database.The rest of the associations found evidence of support in the latest literature,demonstrating the good predictive power of our proposed method.